Deborah Estrin

1
An Architecture for Sensor Networks Directed
Diffusion

• Deborah Estrin
• USC CS Dept and ISI
• In collaboration with
• Co-PIs Ramesh Govindan, John Heidemann
• Diffusion Chalermak Intanagowat, Amit Kumar
• Localized algorithms Jeremy Elson,Satish Kumar,
  Ya Xu, Jerry
  Zhao
• Localization Lew Girod, Nirupama Bulusu
• Distributed robotics Maja Mataric, Gaurav
  Sukhatme, Alberto Cerpa
• For more information estrin_at_isi.edu

2
The long term goal
Embed numerous distributed devices to monitor and
interact with physical world in work-spaces,
hospitals, homes, vehicles, and the environment
(water, soil, air)
Network these devices so that they can coordinate
to perform higher-level tasks. Requires robust
distributed systems of tens of thousands of
devices.
3
Overview of research

• Sensor network challenges
• One approach Directed diffusion
• Basic algorithm
• Initial simulation results (Intanagowat)
• Other interesting localized algorithms in
  progress
• Aggregation (Kumar)
• Adaptive fidelty (Xu)
• Address free architecture, Time synch (Elson)
• Localization (Bulusu, Girod)
• Self-configuration using robotic nodes (Bulusu,
  Cerpa)
• Instrumentation and debugging (Jerry Zhao)

4
The Challenge is Dynamics!

• The physical world is dynamic
• Dynamic operating conditions
• Dynamic availability of resources
• particularly energy!
• Dynamic tasks
• Devices must adapt automatically to the
  environment
• Too many devices for manual configuration
• Environmental conditions are unpredictable
• Unattended and un-tethered operation is key to
  many applications

5
Approach

• Energy is the bottleneck resource
• And communication is a major consumer--avoid
  communication over long distances
• Pre-configuration and global knowledge are not
  applicable
• Achieve desired global behavior through localized
  interactions
• Empirically adapt to observed environment
• Leverage points
• Small-form-factor nodes, densely distributed to
  achieve Physical locality to sensed phenomena
• Application-specific, data-centric networks
• Data processing/aggregation inside the network

6
Directed Diffusion Concepts

• Application-aware communication primitives
• expressed in terms of named data (not in terms of
  the nodes generating or requesting data)
• Consumer of data initiates interest in data with
  certain attributes
• Nodes diffuse the interest towards producers via
  a sequence of local interactions
• This process sets up gradients in the network
  which channel the delivery of data
• Reinforcement and negative reinforcement used to
  converge to efficient distribution
• Intermediate nodes opportunistically fuse
  interests, aggregate, correlate or cache data

7
Illustrating Directed Diffusion
Setting up gradients
Source
Sink
8
Local Behavior Choices

• 1. For propagating interests
• In our example, flood
• More sophisticated behaviors possible e.g. based
  on cached information, GPS
• 2. For setting up gradients
• Highest gradient towards neighbor from whom we
  first heard interest
• Others possible towards neighbor with highest
  energy

• 3. For data transmission
• Different local rules can result in single path
  delivery, striped multi-path delivery, single
  source to multiple sinks and so on.
• 4. For reinforcement
• reinforce one path, or part thereof, based on
  observed losses, delay variances etc.
• other variants inhibit certain paths because
  resource levels are low

9
Initial simulation studies(Intanago, Estrin,
Govindan)
FLOODING

• Compare diffusion to a)flooding, and b)centrally
  computed tree (ideal)
• Key metrics
• total energy consumed per packet delivered
  (indication of network life time)
• average pkt delay

DIFFUSION
CENTRALIZED
CENTRALIZED
DIFFUSION
FLOODING
10
What we really learnt (things we dont usually
showbecause in retrospect they seem so obvious)

• IDLE time dominates energy consumptionneed low
  duty cycle MAC, driven by application.
• With 802.11ish contention protocols you might as
  well just FLOOD
• Easy to get lost in detailed simulations but in
  the wrong region of operation
• Node density, traffic load, stream length, source
  and sink placement, mobility, etc.

11
Exploring Diffusion

• Aggregation
• Adaptive Fidelity
• Implications
• address free architecture
• Need for localization
• Using diffusion
• System health measurements
• Robotic nodes

12
Diffusion based Aggregation(Kumar, Kumar,
Estrin, Heidemann)

• Scaling requires processing of data INSIDE the
  net
• Clustering approach
• Elect cluster head (various promotion criteria)
• Aggregation or Hashing (indirection) to map from
  query to cluster head
• Opportunistic aggregation
• Reinforce (request gradient) proportional to
  aggregatability of incoming data (Amit Kumar)

13
Adaptive Fidelity(Xu, Estrin, Heidemann)

• In densely deployed sensor nets, reduce duty
  cycle engage more nodes when there is activity
  of interest to get higher fidelity
• Adjust node's sleeping time according to the
  number of its neighbors.
• Initial simulations applied to ad hoc routing
• Performance Metric Percentage of survived nodes
  over time.
• The more nodes survive, the longer network
  lifetime

14
Comparison Density factor

• At the left, from top to the bottom Adaptive
  Fidelity, Basic algorithm, regular AODV
• Simulation under 50 nodes, 100 nodes, 150 nodes
• Network lifetime is extended by deploying more
  nodes only with adaptive fidelity algorithm
• Simulations available (ns-2 based)

15
Comparison Traffic Factor

• At the left, from top to the bottom Adaptive
  Fidelity, Basic algorithm, regular AODV
• Simulation under different traffic load 5pkt/s,
  10pkt/s, 15pkt/s, 20pkt/s
• Longer network lifetime in adaptive
• The more traffic load, the greater the advantage
  in terms of network lifetime

16
Adaptive Fidelity conclusions

• Must be applied at application level (because
  just listening/having radio on dominates energy
  dissipation)
• Unfortunate side effect of resource constraints
  is the need to give up (some) layering
• Many open questions as to density thresholds and
  how to design algorithms to exploit it.

17
Implications local addresses?

• Sensor nets maximize usefulness of every bit
• each bit transmitted reduces net lifetime
• cant amortize large headers for low data rates
• underutilized address space is bad
• Still need to identify transmitter
• Reinforcements, Fragmentation
• Use small, random transaction identifiers
  (locally selectedlike multicast addresses)
• Treat identifier collisions as any other loss
• Address-free method can win in networks with
  locality
• simultaneous transactions at any one point is
  much less than in network as a whole

18

• No need for global address assignmentbut how
  inefficient is it?
• AFA optimizes number of bits used per packet
• o Fewer bits less overhead per data bit
• o More bits less contention loss

Efficiency of AFA as a function of local address
size.
19
Implcations Need Localization(Bulusu, Girod)

• Many contexts you cant have GPS on every node
• form factor
• energy
• obstructions
• Beacon architecture
• Signal strength alone problematic/hopeless

• Federated coordinate systems
• Acoustic ranging (client node asks beacons to
  send chirp and monitors time of flight)
• Self-configuring beacon placement using robotic
  nodes

20
Localization is a critical service(Girod)

• Devices take up physical space
• Sufficiently fine-grained spatial coordinates
  provide implicit routing information (e.g.
  directing interests)
• Location is relevant to many applications
• Devices are doing things in the world users need
  to find them inputs and outputs to tasks often
  reference locations
• How can we achieve fine-grained localization?
• Need sensors to measure distance (ranging)
• Time arrivals of 3 requested acoustic signals
  not signal strength
• Relative or Global?
• Relative spatial measurements more accurate
  because observed phenomena are local, shorter
  ranges, etc.
• Global measurements (e.g. GPS) coarser (40m) but
  provide single coordinate system that can be
  exported unambiguously
• Combine global scope of GPS with precision of
  relative sensors fuse local global coordinate
  frames

21
Localization relies on beacons(Bulusu,
Heidemann, Estrin)

• Precision of localization depends on beacon
  density/placement
• Uniform placement not good solution in real
  environments
• Obstacles, walls, etc prevent inference based on
  signal strength/proximity detection
• Self-configuring beacon placement is interesting
  application for robotic nodes
• Given obstacles, unpredictable propagation
  effects, need empirical placement

22
Sensor Network Tomography(Zhao, Govindan, Estrin)

• Continuously updated indication of sensor network
  health
• Useful for
• performance tuning
• adjusting sensing thresholds
• incremental deployment
• refurbishing sections of sensor field with
  additional resources
• self testing
• validating sensor field response to known input

Tomogram indicating connection quality
23
Sensor Network Tomography Key Ideas and
Challenges

• Kinds of tomograms
• network health
• resource-level indicators
• responses to external stimuli
• Can exchange resource health
• during low-level housekeeping functions
• such as radio synchronization
• Key challenge energy-efficiency
• need to aggregate local representations
• algorithms must auto-scale
• outlier indicators are different

24
Self configuring networks using and supporting
robotic nodes(Bulusu, Cerpa, Estrin, Heidemann,
Mataric, Sukhatme)

• Robotics introduces self-mobile nodes and
  adaptively placed nodes
• Self configuring ad hoc networks in the context
  of unpredictable RF environment

• Place nodes for network augmentation or formation
• Place beacons for localization granularity

25
CONCLUSIONS

• Have just scratched the surface
• We need to put more experimental systems in place
  and start living in instrumented environments or
  we risk too many rat-holes and pipe-dreams
• Long-term and High-impact opportunities
• Biological monitoring
• Environmental sensing
• Medical applications based on micro and nano
  scale devices
• In-situ networks for remote exploration